The Signal to Noise ratio is one of the key factors in determining the quality of a video or stills image. A noisy, grainy picture rarely looks as good as a low noise “clean” image. In addition it’s noise in your images will limit how far you can grade them before the picture quality becomes unacceptably poor.
Almost always what you want is the biggest possible signal with the least possible noise. In a video or film camera the signal is the desired image information or in simple terms the picture. While the noise is…. well….. the noise.
Once upon a time, when film cameras were normal for both still photography and film the noise in the pictures came primarily from the grain structure of the film stock. One of the great features of film cameras is that you can actually change that film stock to suit the type of scene that you are shooting. For low light you could use a more sensitive film stock that was actually truly more sensitive to light. However, often a very sensitive film stock will show more noise as the grain of more sensitive film is normally larger.
With video and digital stills cameras however things are quite different. You can’t normally change the sensor in a video camera and it’s the sensor that determines the sensitivity of the camera and it is the sensor that is the source of the majority of the noise.
Modern CMOS video sensors consist of two parts. The light gathering part and the readout part. The size of the pixels on a sensor is one of the key factors in determining the sensitivity and dynamic range. Small pixels are not good at capturing, converting and storing large numbers of photons of light or electrons of electricity. Bigger pixels are much better at this, so big pixels typically mean better sensitivity and a better dynamic range. Each pixel is unique and as a result every pixel on the sensor will perform slightly differently. The signal stored in the pixels is a tiny analog signal that is easily disturbed by stray electric currents and variations in temperature. As a result of the small variations from pixel to pixel, the stray signals and heat, there is a small variation from moment to moment in the signal that comes off the pixel when it is read out and these variations are what we see as noise.
The analog signal from the pixels is passed to a circuit that converts it to a digital signal. The analog to digital conversion process normally includes some form of noise reduction circuitry to help minimise the noise. By carefully mapping the A to D circuity to the signal range the pixels provide, a sensor manufacturer can find the best combination of noise, dynamic and sensitivity. Once the signal has been converted to a digital one, the noise level, sensitivity and dynamic range is more or less locked in and can’t be changed (Some cameras have the ability to use slightly different A to D conversion ranges to help give improved noise levels at different brightness/dynamic ranges).
The bottom line of all this is that with the vast majority of video cameras the noise level is more or less fixed, as is the sensitivity as we can’t actually swap out the sensor.
But wait! I hear you say…. My camera allows me to change the ISO or gain. Well yes it probably does and in both cases, ISO or gain, with a digital video or stills camera what you are changing is the cameras internal signal amplification. You are NOT making the camera more sensitive, you are simply turning up the volume. As anyone with any type of sound system will know, when you turn up the gain you get more hiss. This is because gain makes not only the desired signal bigger but also the noise. As a result adding gain or increasing the ISO is rarely a great thing to do.
So normally we want to use a digital camera at it’s native sensitivity wherever possible. The native sensitivity is where no gain is being added by the camera or 0dB. In ISO, well you need to find out what the native ISO is and be aware that different gamma curves will have different base ISO’s (which is why I prefer to use dB gain as 0dB = native sensitivity, least noise, best dynamic range, no matter what gamma curve).
To get the best possible image we then want to make our signal (picture information) as big as possible. As we can’t swap out the sensor, the only way to do that is to put as much light as possible onto the sensor. Obviously we don’t want to overload the sensor or exceed the limitations of the recording system, but generally the more light you get on the sensor, the better your pictures will be.
As the sensors noise output remains more or less constant, the best signal to noise ratio will be gained when you put a lot of light on the sensor. This generates a very large signal, so the signal becomes big compared to the noise and the noise becomes only a small percentage of the overall image.
If we are unable to get enough light onto the sensor to expose it fully then it is often tempting to add some gain to make the picture brighter. 6dB of gain is the equivalent to 1 stop of exposure. Just like f-stops, each time we go up a stop we are doubling. So adding 6db of gain doubles everything. It makes the picture the equivalent of one stop brighter, but it also doubles the noise. Adding 12dB gain multiplies the noise 4 times, adding 18dB multiplies the noise 8 times.
What if instead of adding gain to make the picture brighter we let 4x more light fall on the sensor (2 stops)? Well the image gets brighter by the equivalent of 2 stops but as we are not adding gain this means the desirable signal, the picture is now going to be the equivalent of 12dB bigger than the noise than it was before we added the 2 stops of light. That’s going to give you a much cleaner looking image.
How do you get more light onto the sensor? There are many ways such as using a faster lens with a larger aperture that will let more light through. Or you could try using a slower shutter speed (I often find it beneficial in low light to use a 1/24th or 1/25th shutter if there is not too much motion to cause the image to become excessively blurred). Then of course you can also add light to your scene by lighting it. It’s very rare to find noisy and grainy night scenes in feature films and that’s because the night scenes normally have well lit foregrounds but keep dark backgrounds to maintain the sensation of night time or darkness. High contrast is the key to good looking night scenes, well lit foregrounds or actors with deep, dark shadows and backgrounds.
The desire to have a good signal to noise ratio is one of the reasons why when shooting in log or raw you want to expose as brightly as you can (while still maintaining consistent exposure from shot to shot, scene to scene). It’s a little bit harder with standard gammas as we have things like the knee or highlight roll off to deal with. Plus the need to have a shot that looks correct straight out of the camera. But at the end of the day the best results are almost always gained when the gain is kept to a minimum (but don’t use negative gain as this can effect the dynamic range) and the amount of light falling on the sensor as high as possible.
In the next article I’ll give you an interesting experiment to try on a PMW-F5, F55 or PMW-FS7 that is very revealing about the way ISO, gain, exposure and noise behaves that will show why exposing log or raw at +1 to +2 stops is so important.
Excellent piece Alister, very succinct.
Drew